Data Science Protocols for Microstructure Engineering

17
Ahmet Cecen, Tony Fast, and Surya R. Kalidindi Data Science Protocols for Microstructure Engineering

Transcript of Data Science Protocols for Microstructure Engineering

Page 1: Data Science Protocols for Microstructure Engineering

Ahmet Cecen, Tony Fast, and Surya R. Kalidindi

Data Science Protocols for Microstructure Engineering

Page 2: Data Science Protocols for Microstructure Engineering

8/18/2014 9:34 PM 2

Dataset Overview

Property Variable

Interfacial Velocity V

Principal Curvatures K1 & K2

Signed Distance Function ϕ

Dataset Components

Mean Curvature K1�K2

2 H

Gaussian Curvature K1 ∗ K2 K

0

- 5

- 10

5

10

A 2D Slice of the Signed Distance Function (ϕ)

Given ϕ, we explore methods to obtain the explicit location of the interface.

Research Question

Is there a spatial correlation within orbetween curvature and velocity at thesolid-liquid interface?

Page 3: Data Science Protocols for Microstructure Engineering

8/18/2014 9:34 PM 3

Extracting the Boundary

0-Threshold Approach

1) Hard Thresholding of the SignDistance Function (ϕ) at 0:

• f � � � 0, ϕ�� � 01, ϕ�� � 0

2) Identification of Boundary Voxels usingPartial Derivatives: in other words, thefirst voxels inside phase 1 areassumed to form the boundary.

ϕ = 0

ϕ = 1

ϕ = -1

ϕ(x)

Page 4: Data Science Protocols for Microstructure Engineering

8/18/2014 9:34 PM 4

Extracting the Boundary

Delta Function Approach

1) Transform -1.5<ϕ<1.5 with DeltaFunction:

2) Smeared out boundary, commonlyused to evaluate surface integrals.

ϕ(x)

Page 5: Data Science Protocols for Microstructure Engineering

Segmentation for Boundary Pixels

The 0-Threshold Boundary Visualized

Page 6: Data Science Protocols for Microstructure Engineering

8/18/2014 9:34 PM 6

Pre-Processing

1) Obtain V and H at theboundary voxels.

2) Trim V at (-1,1), Trim H at(-0.05,0.05).

3) Back Trim V and H toensure that for each Vthere is a correspondingH. (1-to-1)

4) Scale H to (-1,1).

Cleaning Signals

V H

V H

K is cleaned in a similar

fashion.

Page 7: Data Science Protocols for Microstructure Engineering

8/18/2014 9:34 PM 7

Morphology-Curvature Relationships

K1

K2

K<0

K=0

K>0

Gaussian Curvature – Morphology RelationshipPrincipal Curvatures – Morphology Relationship

� � � ∙ �2 where � is the surface normal.

Mean Curvature – Morphology Relationship

Page 8: Data Science Protocols for Microstructure Engineering

8/18/2014 9:34 PM 8

H – V Full Signal Cross Correlation

YZ – Central Slice

Page 9: Data Science Protocols for Microstructure Engineering

8/18/2014 9:34 PM 9

K – V Full Signal Cross Correlation

YZ – Central Slice

Page 10: Data Science Protocols for Microstructure Engineering

8/18/2014 9:34 PM 10

H – V Partial Signal Cross Correlation

Do high magnitudes of Mean Curvature (H) show a correlation with low magnitudes of Interfacial Velocity (V)?

YZ – Central Slice

Page 11: Data Science Protocols for Microstructure Engineering

e-Collaboration Science

� Identifying the specific incentives andimpediments for participation of cross-disciplinary team members

� designing and deploying suitable softwareenvironments that dramatically improve theefficacy of collaborations

� collaborations largely facilitated by the internet,cloud services, and information sciences

Productivity enhancement of cross-disciplinary teams is the main motivator

for e-collaborations

Page 12: Data Science Protocols for Microstructure Engineering

Advantages of e-Collaborations• Conventional collaborations are structured as

sequential tasks with synchronous exchange ofdata and results

• e-Collaborations can be designed forasynchronous and much more intimate exchangeof data and results while engaging much largergroups

• e-Collaborations allow for iterative workflowsbetween team members in a highly acceleratedmanner.

R1R2 R3

Page 13: Data Science Protocols for Microstructure Engineering

Innovation StackM

ater

ials

Des

ign

an

d

Dev

elo

pm

ent

Presentation

Data Processing

Middleware/Storage

Hardware

Critical NeedRapid

Dissemination of Research Results Using Templated

Presentation Layers

Page 14: Data Science Protocols for Microstructure Engineering

Cloud Services for Sharing• Documents

– Drive, Slideshare, Dropbox, ShareLatex, Mendeley• Data

– Figshare, Dropbox, Globus, FTP• Images

– Flickr, Plot.ly• Codes

– Github, SourceForge, BitBucket, Gitorious

e-Collaborations demand interactive environmentsthat allow discussions, rich annotations, iterativeworkflows, and sharing of expertise

Page 15: Data Science Protocols for Microstructure Engineering

Core Idea: Mini-Research Sites• Automatically launch and populate templated websites with

ongoing (intermediate) research results (including codes,reports, images, data) to selectively disseminate researchprogress to team members

• Advantages� e-portfolio of your research results that can be accessed

on your smartphone� e-portfolio of the research productivity of the entire team� Organized content and figures for publication� e-recording of provenance� Automated extraction of metadata� e-recording of integrated workflows

Page 16: Data Science Protocols for Microstructure Engineering

Presentation Layer• Data � Template � Presentation Layer

� Presentation Layers are some combination of websites, reports, powerpoint slides, figures

� Strong focus on visual communication of centralized research results

• Web templates will be tailored for specific projects (material systems, simulation, experiment, analytics) to present the underlying content in the most easy to understand formats

• Blog-oriented approach to research science– Share success, challenges, and failures

Page 17: Data Science Protocols for Microstructure Engineering

Sites

• Image Identification - http://tonyfast.com/d3-data-browse/• Spatial Data Analytics Viz -

http://tonyfast.com/statviz/2014/07/30/window-500.html• Molecular Dynamics - http://tonyfast.com/Atomic-

Positions/• Titanium - http://tonyfast.com/Titanium/• GOALI(Latex)- http://tonyfast.com/nsf-goali/• Surface Roughness - http://tonyfast.com/Aluminum-

Surface-Roughness/• Spatial Statistics -

http://tonyfast.com/SpatialStatisticsFFT/• PyMKS http://openmaterials.github.io/pymks/index.html